INTERACTIVE PHYSICAL DESIGN TUNING

Abstract

An architecture for providing interactive sessions for physical database design is described, allowing users to readily try different options, identify problems, and obtain physical designs in a flexible way. Embodiments based on a .NET assembly and modifications to a database management system (DBMS) are also described.

Description

BACKGROUND

Automated physical design tuning involves a database management system (DBMS) recommending a set of physical structures that increase the performance of an underlying database. Physical design has been formulated as a problem statement, traditionally: Given a workload W and a storage budget B, find the set of physical structures, or configuration, that fits in B and results in the lowest execution cost for W. Most modern commercial DBMS's have some facilities for automated design tuning. In general, however, it has not been possible to include in the tuning process information beyond the basic information of the design tuning problem statement.

For instance, it has not been possible to tune a given workload for maximum performance under a storage constraint while at the same time ensuring that no query degrades by more than 10% with respect to the original configuration. As another example, it has not been possible to enforce that the clustered index on a table T cannot be defined over certain columns of T that would introduce hot-spots (without specifying which of the remaining columns should be chosen). As yet another example, in order to decrease contention during query processing, there is no way to avoid any single column from a table from appearing in more than, say, three indexes (the more indexes a column appears in, the more contention arises due to exclusive locks during updates). While some new approaches allow more flexibility in the specification of a physical design tuning problem, existing solutions require that the whole specification to be provided upfront, without possibility of interaction.

Described herein are techniques for flexible and interactive physical design tuning.

SUMMARY

The following summary is included only to introduce some concepts discussed in the Detailed Description below. This summary is not comprehensive and is not intended to delineate the scope of the claimed subject matter, which is set forth by the claims presented at the end.

An architecture for providing interactive sessions for physical database design is described, allowing users to readily try different options, identify problems, and obtain physical designs in a flexible way. Embodiments based on a .NET assembly and modifications to a database management system (DBMS) are also described.

Many of the attendant features will be explained below with reference to the following detailed description considered in connection with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

The present description will be better understood from the following detailed description read in light of the accompanying drawings, wherein like reference numerals are used to designate like parts in the accompanying description.

FIG. 1 shows a physical design tuning tool.

FIG. 2 shows a layered architecture for physical design tuning.

FIG. 3 shows a transcript of an example interactive tuning session.

FIG. 4 shows an XML representation of a portion of a simplified version of a hierarchy exposed by a provider.

FIG. 5 shows a transcript of an example interactive tuning session using a provider.

FIG. 6 shows an example visualization.

FIG. 7 shows an example PowerShell session with two example cmdlets.

FIG. 8 shows an example script.

FIGS. 9-11 show a sample interactive physical database design tuning session.

FIG. 12 shows a process of interactive design tuning.

FIG. 13 shows a diagram illustrating user interaction possible for design tuning.

DETAILED DESCRIPTION

Overview

Embodiments discussed below relate to interactive physical design tuning of databases. For background, FIG. 1 shows a physical design tuning tool 130. The tuning tool 130 is an application used by a DBA to explore alternate configurations of a database 131 managed by a DBMS. Different storage boundaries 132 (e.g., memory limits) and physical configurations and workloads 134 (sets of queries) can be tested for performance results. The tuning tool 130 may function as a database client that communicates with the DBMS through an interface (an API provided by the database) to submit “what-if” scenarios that the DBMS explores, tests, and provides feedback on without necessarily executing the queries. In other words, the DBMS 102 can give feedback on how well a query optimizes or performs according to a possible configuration. As a result, the tuning tool 130 may output some configuration 136 (e.g., some set of indexes I1 . . . Im) deemed to be ideal for the given boundaries 132 and workload 134. Generally, the tuning tool 130 may test many configurations for a query before deciding which configuration is optimal. Regarding what-if facilities or instrumentation of a DBMS, description is provided elsewhere, for example, see “Self-Tuning Database Systems: A Decade of Progress”, VLDB '07, ACM 978-1-59593-649-Mar. 7, 2009.

The scenarios mentioned in the Background above show that the state-of-the-art techniques for physical design tuning are inflexible. Referring to FIG. 1, the typical approach has been to specify a simple scenario, submit same to the DBMS, and receive a recommended configuration. Flexible and interactive refinement has not been possible. The inventors have observed that goals or constraints of tuning the physical design of a database system are often not fully specified upfront, but instead become apparent through experimentation, feedback, analysis, etc. Current physical design tools are monolithic, expose tuning options that are set at the beginning, and generate, without further user input, a final configuration to deploy into a production system.

Embodiments described below shift the design approach and allow tuning sessions to be highly interactive. Current monolithic architectures in physical design tools force users to specify the whole problem upfront and prevent users from making changes a posteriori or in general interacting with the system. Explanation will begin with description of an architecture for interactive sessions, followed by a review of Windows PowerShell as an infrastructure component that can support the architecture. Explanation will proceed with description of interactive tuning processes, followed by presentation of illustrative examples.

Layered Architecture for Physical Design Tuning

FIG. 2 shows a layered architecture 148 for physical design tuning. A Core DBMS layer 150 is a lowest layer within the DBMS (e.g., SQL Server 152) and provides native support for operations such as what-if optimization, which may be facilitated in party by a SQL Server Optimizer 154, for example. What-if optimization support in a DBMS is described in other sources, such as “Configuration-parametric query optimization for physical design tuning” (In Proceedings of the ACM International Conference on Management of Data (SIGMOD), 2008, and “Autoadmin ‘What-if’ index analysis utility” (In Proceedings of the ACM International Conference on Management of Data (SIGMOD), 1998). This layer provides what-if optimization (i.e., the ability to obtain expected costs of queries for varying physical designs without the requirement that the physical designs being implemented). Additionally, this layer enables rich information after regular optimization (see low layer API), such as sets of access path requests that can be later used to understand the set of possible indexes that could be useful for a given query. It does so by piggybacking on top of regular optimization and identifying access path requests. It may not actually expose functions, but instead may enable functionality that can be exposed in the low-level APIs.

A low-level API layer 156 may expose, in formats that are simple to consume (e.g., XML), the functionality of the Core DBMS layer 150 (and also the DBMS itself). As an example, they may expose primitives to manipulate a what-if mode of the DBMS and also may expose rich explain modes which, after optimizing queries, surface optimization information use at higher levels of the DBMS. The explain mode may provide useful information about the optimization of a query, such as the final plan obtained by the optimizer, cardinality estimates for intermediate results, access path requests, etc. (it may be thought of as an extension to existing modes in relational systems, such as showplans in Microsoft SQL Server). The low level API layer 156 may also encapsulate existing DBMS functionality, such as mechanisms that monitor and gather workloads.

A High-level API layer 158 if provided to facilitate access to the lower level APIs 156 and Core DBMS layer 150. Physical design tools were previously built on top of the low-level APIs only exposed a rigid functionality (e.g., point to a workload, set the storage constraint, and optimize). The high-level API layer 158 exposes the internal representations and mechanisms in a modular way. Basic concepts such as queries, indexes, databases, tables, and access-path requests are exposed as instantiable classes. In addition to these data structures, the high-level API layer 158 exposes composable and simple algorithms sometimes found in previous tuning tools. For instance, this layer may expose mechanisms to merge two indexes, or to obtain the best set of indexes for a single query. These primitive data structures and algorithms are not necessarily meant to be consumed by DBAs, but instead provide a foundational abstraction for applications to be built on top, as explained next. In one embodiment, described later, the high-level API layer 158 may be implemented as a .NET assembly 160, which is executed by a .NET VM 162 (Virtual Machine), sometimes called a managed code environment.

Front-ends 164 are based on both the low-level APIs 156 and high-level APIs 158 and deliver functionality to end users. One example of a front-end 164 is an interactive scripting platform to interact with physical database designs. The scripting language understands and works with the data structures and algorithms exposed by the underlying layers and allows users to write interactive scripts to tune the physical design of a database. Common tasks, such as minimizing the cost for a single storage constraint (or other functionality provided by previous physical design tools), can be implemented as pre-existing scripts that can be accessed using graphical user interfaces by relatively inexperienced DBAs.

As mentioned, a front-end 164 can be implemented by a scripting environment. For example, Windows Powershell 166 (tm), available from Microsoft Corporation is a scripting language that can be used as a front-end 164 in the architecture. A prototype implementation of the architecture using Windows Powershell 166 will also be described.

Windows Powershell

Windows PowerShell is an interactive, extensible scripting language that integrates with the Microsoft .NET Framework. It provides an environment to perform administrative tasks by execution of cmdlets (i.e., commandlets, which are basic operations), scripts (which are composition of cmdlets), stand-alone applications, or by directly instantiating regular .NET classes. The main features of Windows PowerShell include tight integration with .NET, strict naming conventions, object pipelines, and data providers.

Windows PowerShell integrates with the .NET framework and leverages the .NET framework to represent data. Windows PowerShell understands .NET classes natively, as illustrated below. Thus, new classes written in the .NET framework are easily available as first-class citizens in Windows PowerShell.

Windows PowerShell uses strict naming conventions. Cmdlets in Windows PowerShell follow a verb-noun naming convention, and parameters are passed in a unified manner. Some examples of such built-in cmdlets are Start-Service, which starts an OS (operating system) service in the current machine, Get-Process, which returns a list of processes currently executing, Clear-Host, which clears the screen, and Get-ChildItem which, if located in a file system directory, returns all its subdirectories or files. There are also aliases for the common cmdlets.

PowerShell also provides facilities to construct object pipelines. Similar to Unix shells, cmdlets can be pipelined using the “|” operator. However, unlike Unix shells, which typically pipeline strings, Windows PowerShell pipelines .NET objects. For instance, the script:

> Get-Process | Sort-Object -Property Handles -Desc |
Select-Object -first 5 | Stop-Process

obtains the list of all running processes, pipes the result (which is a list of System.Diagnostics.Process.NET objects) to the Sort-Object cmdlet, which understands the semantics of the objects and sorts them by the property Handles in descending order. In turn, the result of this cmdlet (i.e., an ordered list of processes) is passed to the Select-Object cmdlet, which takes the first five processes and passes them to the next cmdlet in the pipeline, Stop-Process, which terminates them. The following script returns the number of lines that contains the word “constraint” in any LATEX file in the current directory that is below 100,000 bytes long:

> Get-ChildItem -Path *.tex | Where-Object -FilterScript { $ .Length -It
100000 } | Foreach-Object -Process { Get-Content $ | Select-String
constraint } | Measure-Object Count : 404

which gets all files in the current path that have a “tex” extension and keeps only those that are smaller than 100,000 bytes. Then, each file is processed by first getting its content (which returns a list of string .NET classes), selecting only those that contain the work constraints. The combined result of this subscript is a list of strings, which is measured and the count is returned. To shorten a script, aliases (e.g., Get-ChildItem becomes “dir”, Where-Object becomes “?”, Foreach-Object becomes %), and positional cmdlet parameters can be used. For instance it is not necessary to explicitly write—Path after dir. An equivalent script is shown below:

> dir *.tex | ? { $ .Length -It 100000 } | % { gc $ | Select-String
constraint } | measure Count : 404

PowerShell has the ability to expose hierarchical data models by means of data providers, which are then accessed and manipulated using a common set of cmdlets. As an example, the file system is one such provider. When situated in some node inside the file system provider, Get-ChildItem can be used to obtain the subdirectories or files in the current location, access contents of elements using Get-Content, and navigate the provider using Set-Location (aliased as cd). However, Windows PowerShell natively exposes the registry and the environment variables as providers. There also are third party providers that give a unified surface to access, query, and modify Active Directory, SharePoint and SQL Server, among others.

The next section describes how take advantage of the different features of Windows PowerShell to provide an interactive experience for physical design tuning.

Interactive Physical Design Tuning

A prototype implementation that enables interactive physical design tuning sessions will now be described. The architecture of this implementation is described first, followed by discussion of examples of how the implementation can be used. FIG. 2 shows how the different layers of architecture 148 map to the implementation. An implementation of each layer of the architecture 148 will be described in detail.

Low-Level APIs

The Core DBMS 150 and Low-level APIs 156 are implemented by instrumenting a database server, for instance Microsoft SQL Server 152. Some components (e.g., what-if optimization) are already part of this particular database server, while others (e.g., access-path request interception) were added.

High-Level APIs

The high-level API layer 158 is implemented by introducing a new .NET assembly 160 that encapsulates and exposes classes and algorithms relevant to physical design tuning. Among the classes that the assembly exposes are Database, Table, Index, Column, Query, Configuration, and Request classes. These are rich in functionality, so for instance the Index class may have methods that return merged and reduced indexes and methods that create hypothetical versions of the index in the database. The Query class may have methods that evaluate (optimize) it under a given configuration, and methods that return its set of access-path requests.

Additionally, as part of the .NET assembly 160, a sophisticated caching mechanism may be built to avoid optimizing the same query multiple times in the database server. Instead, each query remembers previous optimizations and, if asked again to optimize itself with a previously seen configuration, it returns the cached values without doing the expensive work again. Because these classes are exposed in an assembly, the definitions thereof can be loaded directly into Windows PowerShell which may be used to explore, in interactive form, the physical design of a database, as illustrated in FIG. 3.

FIG. 3 shows a transcript 180 of an example interactive tuning session. First, a user created 182 a new database object; an object implemented by the .NET assembly but instantiated via user input typed at the PowerShell command prompt (“>”). Then, the user explored the database object by typing its variable name 184, in response to which various information about the corresponding data is displayed (such information having been obtained from “nicolasb02” DBMS hosting the “tpch01g” database). Specific information about tables of the database is then displayed by invoking 186 appropriate properties of the $db database object. A new query object is then instantiated 188, and then the user evaluates the query 190 according to the “base” or default configuration of the database.

Front-End

While the example above is useful, call the .NET methods directly can be inefficient. Also, using such methods directly may be a time consuming way to accomplish tuning a database design. Using the capabilities of Windows PowerShell, functionally such as a provider, visualizations, cmdlets, and scripts can be used. PowerShell providers are .NET programs that allow a user to work with data stores as though they were mounted drives or file systems, which simplifies accessing external data outside the PowerShell environment. A PowerShell provider can be implemented that exposes the information about a tuning session in a hierarchical and intuitive object model. FIG. 4 shows an XML representation 210 of a portion of a simplified version of a hierarchy exposed by a provider. By using this provider, the state of a tuning session can be easily manipulated and navigated as shown in FIG. 5. FIG. 5 shows a transcript 230 of an example interactive tuning session using a provider. Note that file system style commands such as “cd” and “dir” can be used to navigate and view database information, tuning information, configuration information, and so on.

FIG. 6 shows an example visualization 250. Use of a composable, interactive script language can allow inclusion of third-party add-ins that offer specific functionality. For example, the PowerGadgets web site provides simple cmdlets to display data graphically. One such cmdlet is Out-Chart, of which visualization 250 is an instance. The Out-Chart cmdlet displays a chart of the data that is pipelined in. The example visualization 250 graphically displays the relative sizes of all tables in a database by using the command 252.

In addition to a provider, the bare .NET classes and methods of the .NET assembly may be provided with composable cmdlets. FIG. 7 shows an example PowerShell session with two example cmdlets 270 and 272.

Scripts are another feature of the implementation. FIG. 8 shows an example script 290. The script is a simplified implementation of a common operation called refinement, in which a given input configuration is repeatedly “relaxed” via merging and reduction operations until it fits in the available storage, so that the expected cost of the resulting configuration is as low as possible. At each iteration, all possible transformations are calculated (using a cmdlet) and the one that is expected to result in the smaller execution cost is obtained. This process is repeated until a valid configuration is reached:

Other common algorithms may be similarly implemented, such as the relaxation based tuning approach in “Automatic physical database tuning: A relaxation-based approach” (N. Bruno and S. Chaudhuri., In Proceedings of the ACM International Conference on Management of Data (SIGMOD), 2005). One embodiment implements a version that handles constraint language, described in “Constrained physical design tuning” (N. Bruno and S. Chaudhuri, In Proceedings of the International Conference on Very Large Databases (VLDB), 2008). This script is called TuneConstrained-Workload and takes as inputs a workload, a timeout, and a set of constraints. Such a script may be implemented by using the .NET classes exported by the high-level APIs and may be implemented as a PowerShell script in fewer than 100 lines of code.

A Sample Interactive Tuning Session

FIGS. 9-11 show a sample interactive physical database design tuning session 310. Session 310 illustrates an interactive approach that provides flexibility and control during physical design tuning. The example uses the provider, cmdlets and scripts described earlier, as well as additional visualizations, among others. Advanced DBAs or other users may create their own scripts to further customize the physical design tuning experience. Moreover, native PowerShell features, such as remoting (which allows users to execute commands in other machines) or eventing and automation can surely complement tuning scripts and provide added flexibility. The comments in session 310 explain the actions taken by the user.

FIG. 12 shows a process of interactive design tuning. The steps of FIG. 12 can occur in any arbitrary order; the order shown is for illustration only. Using an implementation of the architecture described above, a user may configure 330 a scenario by selecting workloads, setting parameters such as space constraint, and others. Eventually the user invokes a command (e.g., “Refine-Configuration”) that communicates 332 the current scenario or configuration to a low-level API algorithm implemented by a DBMS. The DBMS may load 334 a workload, set parameters, and analyze the workload to find an optimal physical design, indicia of which (e.g., graphs, text) is outputted 336 to the user. The user may decide 338 to perform additional steps, again, in arbitrary order.

FIG. 13 shows a diagram illustrating the free-form manner of user interaction possible for design tuning. The user continuously enters user input 360, which may be specification commands 362 that specify the scenario such as a configuration, a workload or queries, databases, etc. Generally, specification commands are handled by the shell, front-end, high-level API (e.g., assembly), etc., rather than the DBMS. The user may also provide user input 360 in the form of commands 364 that invoke what-if functionality and other design functionality provided by the DBMS via the low-level API and/or core DBMS. While a command line shell/interpreter implementation has been described, graphical user interfaces can also be used, either as programs directly interfacing with the low-level API and high-level API, or by interfacing with another front-end such as a character-based shell environment.

Conclusion

Embodiments and features discussed above can be realized in the form of information stored in volatile or non-volatile computer or device readable media. This is deemed to include at least media such as optical storage (e.g., CD-ROM), magnetic media, flash ROM, or any current or future means of storing digital information. The stored information can be in the form of machine executable instructions (e.g., compiled executable binary code), source code, bytecode, or any other information that can be used to enable or configure computing devices to perform the various embodiments discussed above. This is also deemed to include at least volatile memory such as RAM and/or virtual memory storing information such as CPU instructions during execution of a program carrying out an embodiment, as well as non-volatile media storing information that allows a program or executable to be loaded and executed. The embodiments and features can be performed on any type of computing device, including portable devices, workstations, servers, mobile wireless devices, and so on.

Claims

1. A computer-implemented method to allow a user to interactively explore a physical database design, the method comprising:

accessing a database management system (DBMS) including a database engine;

allowing a user to interactively input commands in arbitrary order, the commands including: specification commands specifying a database of the DBMS; specification commands specifying one or more queries directed to the database; specification commands specifying one or more physical configurations of the database, a specified physical configuration comprising at least set of indices of tables of the database; and execution commands obtaining performance and/or optimization information from the DBMS, the performance and/or optimization information generated by an API of the DBMS that provides access to what-if functions of the DBMS that analyze query execution without fully executing queries.

2. A computer-implemented method according to claim 1, wherein the commands are entered via a command line interpreter that receives the interactively inputted commands and which allows the user to create arbitrary variables and objects to be used as parameters of the commands.

3. A computer-implemented method according to claim 2, further comprising providing an interface between the DBMS and the command line interpreter, wherein the command line interpreter communicates with objects of the module and the module communicates with an API of the DBMS.

4. A computer-implemented method according to claim 1, the DBMS comprising a query optimizer, a low-level API communicating with the query optimizer, the device comprising an assembly executed by a managed code environment, the assembly communicating with the low level API, the device further comprising a front-end program to which the user enters the commands.

5. A computer-implemented method according to claim 4, wherein in response to some of the commands the program creates or alters objects that are instances of classes implemented by the assembly without communicating with the DBMS.

6. A computer-implemented method according to claim 5, wherein in response to some of the commands the program communicates one or more properties of the objects of the assembly via the low-level API to the DBMS and in response receives the performance and/or optimization information from the DBMS.

7. A computer-implemented method according to claim 1, wherein the commands are entered via a program separate from the database, when an execution command is executed by the user, information entered by the specification commands is used to obtain the performance and/or optimization information from the DBMS, and when an execution command is executed the information entered by the specification commands continues to be maintained by the program and after a first execution command is executed and performance and/or optimization information is obtained by the program, a second execution command can be entered at any arbitrary time using the stored information of the first execution command.

8. A computer-implemented method according to claim 7, wherein the program comprises a shell environment with general programming features that allow the user to arbitrarily name variables and assign values to the variables.

9. One or more computer-readable storage media storing information to enable a computing device to perform a process, the process comprising:

establishing a connection between a component and a DBMS, the component configured to enable a user to interact with a tuning API of a DBMS, the tuning API receiving hypothetical scenarios and submitting them to the DBMS and returning corresponding physical database design configurations generated by the DBMS, the physical database design configurations comprising configurations of a database managed by the DBMS, wherein the interactive component is configured to allow the user to interactively build and modify hypothetical scenarios and at any arbitrary time while doing so allows the user to submit one of the hypothetical scenarios to the DBMS via the tuning API, wherein the interactive component maintains the hypothetical scenarios between submissions of same via the tuning API.

10. One or more computer-readable storage media according to claim 9, wherein the interactive component comprises an intermediary between an application program and the DBMS, and the user builds and modifies the hypothetical scenarios by interacting with the application program.

11. One or more computer-readable storage media according to claim 9, wherein the application program includes an implementation of a scripting language, the user interacting with the application program to formulate commands in the scripting language.

12. One or more computer-readable storage media according to claim 9, wherein the component exposes objects that represent structures of the database and exposes physical design tuning functions, the commands in the scripting language referencing the objects and the design tuning functions exposed by the component.

13. One or more computer-readable storage media according to claim 9 wherein the user interaction is with a programming environment in which the user can create and manipulate arbitrary variables that are used to define the arbitrary scenarios.

14. One or more computer-readable storage media according to claim 13, wherein the exposed objects include objects representing databases, tables, configurations comprised of at least indices, and queries.

15. One or more computer-readable storage media according to claim 9, wherein the physical database design configurations are generated by optimization analysis performed by the DBMS.

16. One or more computer-readable storage media storing information configured to enable a computing device to perform a process, the process performed by one or more components separate from and in communication with a DBMS, the process comprising:

providing commands invoked by a user to instantiate objects and designate variables pointing to the objects;

providing what-if commands executable by the user to invoke what-if analysis of a DBMS; and

providing an interaction component that receives interactive input from a user and provides corresponding output to a display device, wherein the interactive input allows the user to invoke the what-if commands which reference the objects.

17. One or more computer-readable storage media according to claim 16, further comprising: providing math commands for arbitrary math computations and text commands for arbitrary text computations, and wherein the interaction component allows the user to construct arbitrary compound commands by interactively specifying, for a compound command, a what-if command, a math or text command, where the compound command references or provides output to one of the variables, and where output of one the compound command's commands is provided as input to another of the compound command's commands.

18. One or more computer-readable storage media according to claim 16, wherein at least one of the components comprises an assembly executed by a managed code environment, the one of the components interfacing with one or more APIs of the DBMS and implementing classes representing database structures of which the objects instantiated by the user are instances.

19. One or more computer-readable storage media according to claim 16, wherein the user invokes a command to specify a workload, and the user repeatedly invokes commands to refine a hypothetical configuration of the database and repeatedly invokes the what-if commands between refinements of the hypothetical configuration of the database.

20. One or more computer-readable storage media according to claim 16, further comprising providing a visualization command that is included as part of the compound command and receives output of one of the commands of the compound commands and generates corresponding graphical representation of the output.